We present a fundamentally different approach to classifying traffic flows according the applications that generate them. In contrast previous methods, our is based on observing and identifying patterns of host behavior at transport layer. analyze these three levels increasing detail (i) social, (ii) functional (iii) application level. This multilevel looking flow probably most important contribution this paper. Furthermore, has two features. First, it operates in dark, having (a) no access...

The shared nature of the network in today's multi-tenant datacenters implies that performance for tenants can vary significantly. This applies to both production and cloud environments. Network variability hurts application which makes tenant costs unpredictable causes provider revenue loss. Motivated by these factors, this paper case extending tenant-provider interface explicitly account network. We argue be achieved providing with a virtual connecting their compute instances. To effect,...

Since the emergence of peer-to-peer (P2P) networking in late '90s, P2P applications have multiplied, evolved and established themselves as leading `growth app' Internet traffic workload. In contrast to first-generation networks which used well-defined port numbers, current ability disguise their existence through use arbitrary ports. As a result, reliable estimates require examination packet payload, methodological landmine from legal, privacy, technical, logistic, fiscal perspectives....

The soft real-time nature of large scale web applications in today's datacenters, combined with their distributed workflow, leads to deadlines being associated the datacenter application traffic. A network flow is useful, and contributes throughput operator revenue if, only it completes within its deadline. Today's transport pro- tocols (TCP included), given Internet origins, are agnostic such deadlines. Instead, they strive share resources fairly. We show that this can hurt performance.

We examine the fundamental properties that determine basic performance metrics for opportunistic communications. first consider distribution of inter-contact times between mobile devices. Using a diverse set measured mobility traces, we find as an invariant property there is characteristic time, order half day, beyond which decays exponentially. Up to this value, in many cases follows power law, shown recent work. This powerlaw finding was previously used support hypothesis time has law...

Recent reports in the popular media suggest a significant decrease peer-to-peer (P2P) file-sharing traffic, attributed to public's response legal threats. Have we reached end of P2P revolution? In pursuit legitimate data verify this hypothesis, paper, embark on more accurate measurement effort traffic at link level. contrast previous efforts, introduce two novel elements our methodology. First, measure all known protocols. Second, go beyond "known port" limitation by reverse engineering...

Since the identification of long-range dependence in network traffic ten years ago, its consistent appearance across numerous measurement studies has largely discredited Poisson-based models. However, since that original data set was collected, both link speeds and number Internet-connected hosts have increased by more than three orders magnitude. Thus, we now revisit Poisson assumption, studying a combination historical traces new measurements obtained from major backbone belonging to Tier...

We present a fundamentally different approach to classifying traffic flows according the applications that generate them. In contrast previous methods, our is based on observing and identifying patterns of host behavior at transport layer. analyze these three levels increasing detail (i) social, (ii) functional (iii) application level. This multilevel looking flow probably most important contribution this paper. Furthermore, has two features. First, it operates in dark , having (a) no access...

Recently, peer-to-peer (P2P) networks have emerged as an attractive solution to enable large-scale content distribution without requiring major infrastructure investments. While such P2P solutions appear highly beneficial for providers and end-users, there seems be a growing concern among Internet Service Providers (ISPs) that now need support the cost. In this work, we explore potential impact of future file delivery mechanisms seen from three different perspectives: i) provider, ii) ISPs,...

We examine the fundamental properties that determine basic performance metrics for opportunistic communications. first consider distribution of intercontact times between mobile devices. Using a diverse set measured mobility traces, we find as an invariant property there is characteristic time, order half day, beyond which decays exponentially. Up to this value, in many cases follows power law, shown recent work. This law finding was previously used support hypothesis time has tail, and...

Many data center applications perform rich and complex tasks (e.g., executing a search query or generating user's news-feed). From network perspective, these typically comprise multiple flows, which traverse different parts of the at potentially times. Most resource allocation schemes, however, treat all flows in isolation -- rather than as part task therefore only optimize flow-level metrics.

Self-similarity and scaling phenomena have dominated Internet traffic analysis for the past decade. With identification of long-range dependence (LRD) in network traffic, research community has undergone a mental shift from Poisson memory-less processes to LRD bursty processes. Despite its widespread use, though, is hindered by difficulty actually identifying estimating parameters unambiguously. The authors outline findings explore current lack accuracy robustness estimation. In addition,...

The shared nature of the network in today's multi-tenant datacenters implies that performance for tenants can vary significantly. This applies to both production and cloud environments. Network variability hurts application which makes tenant costs unpredictable causes provider revenue loss. Motivated by these factors, this paper case extending tenant-provider interface explicitly account network. We argue be achieved providing with a virtual connecting their compute instances. To effect,...

The soft real-time nature of large scale web applications in today's datacenters, combined with their distributed workflow, leads to deadlines being associated the datacenter application traffic. A network flow is useful, and contributes throughput operator revenue if, only it completes within its deadline. Today's transport pro- tocols (TCP included), given Internet origins, are agnostic such deadlines. Instead, they strive share resources fairly. We show that this can hurt performance....

In data centers, the IO path to storage is long and complex. It comprises many layers or "stages" with opaque interfaces between them. This makes it hard enforce end-to-end policies that dictate a flow's performance (e.g., guarantee tenant's bandwidth) routing route an untrusted VM's traffic through sanitization middlebox). These require differentiation along flow global visibility at control plane. We design IOFlow, architecture uses logically centralized plane enable high-level policies....